Make reboot work for halted xenapi instances

Fixes bug 1022199

This also will catch exceptions and make sure to reset the task_state,
while still generating an instance fault.

Change-Id: I122a1422b8e5731bc484414736ab44e60d4c9830

4 files changed, 69 insertions, 7 deletions

diff --git a/nova/compute/manager.py b/nova/compute/manager.py
index afd422253..aacfdf7a7 100644
--- a/nova/compute/manager.py
+++ b/nova/compute/manager.py
@@ -959,8 +959,15 @@ class ComputeManager(manager.SchedulerDependentManager):
                      context=context, instance_uuid=instance_uuid)
 
         network_info = self._get_instance_nw_info(context, instance)
-        self.driver.reboot(instance, self._legacy_nw_info(network_info),
-                           reboot_type)
+        try:
+            self.driver.reboot(instance, self._legacy_nw_info(network_info),
+                    reboot_type)
+        except Exception, exc:
+            LOG.error(_('Cannot reboot instance: %(exc)s'), locals(),
+                    context=context, instance_uuid=instance_uuid)
+            self.add_instance_fault_from_exc(context, instance_uuid, exc,
+                    sys.exc_info())
+            # Fall through and reset task_state to None
 
         current_power_state = self._get_power_state(context, instance)
         self._instance_update(context,
diff --git a/nova/tests/test_xenapi.py b/nova/tests/test_xenapi.py
index 3138ef0b9..aeacf1fa6 100644
--- a/nova/tests/test_xenapi.py
+++ b/nova/tests/test_xenapi.py
@@ -807,6 +807,33 @@ class XenAPIVMTestCase(stubs.XenAPITestBase):
         conn.finish_revert_migration(instance, None)
         self.assertTrue(conn._vmops.finish_revert_migration_called)
 
+    def test_reboot_hard(self):
+        instance = self._create_instance()
+        conn = xenapi_conn.XenAPIDriver(False)
+        conn.reboot(instance, None, "HARD")
+
+    def test_reboot_soft(self):
+        instance = self._create_instance()
+        conn = xenapi_conn.XenAPIDriver(False)
+        conn.reboot(instance, None, "SOFT")
+
+    def test_reboot_halted(self):
+        session = xenapi_conn.XenAPISession('test_url', 'root', 'test_pass')
+        instance = self._create_instance(spawn=False)
+        conn = xenapi_conn.XenAPIDriver(False)
+        xenapi_fake.create_vm(instance.name, 'Halted')
+        conn.reboot(instance, None, "SOFT")
+        vm_ref = vm_utils.lookup(session, instance.name)
+        vm = xenapi_fake.get_record('VM', vm_ref)
+        self.assertEquals(vm['power_state'], 'Running')
+
+    def test_reboot_unknown_state(self):
+        instance = self._create_instance(spawn=False)
+        conn = xenapi_conn.XenAPIDriver(False)
+        xenapi_fake.create_vm(instance.name, 'Unknown')
+        self.assertRaises(xenapi_fake.Failure, conn.reboot, instance,
+                None, "SOFT")
+
     def _create_instance(self, instance_id=1, spawn=True):
         """Creates and spawns a test instance."""
         instance_values = {
@@ -938,6 +965,14 @@ class XenAPIMigrateInstance(stubs.XenAPITestBase):
         conn.migrate_disk_and_power_off(self.context, instance,
                                         '127.0.0.1', instance_type, None)
 
+    def test_migrate_disk_and_power_off(self):
+        instance = db.instance_create(self.context, self.instance_values)
+        xenapi_fake.create_vm(instance.name, 'Running')
+        instance_type = db.instance_type_get_by_name(self.context, 'm1.large')
+        conn = xenapi_conn.XenAPIDriver(False)
+        conn.migrate_disk_and_power_off(self.context, instance,
+                                        '127.0.0.1', instance_type, None)
+
     def test_migrate_disk_and_power_off_passes_exceptions(self):
         instance = db.instance_create(self.context, self.instance_values)
         xenapi_fake.create_vm(instance.name, 'Running')
diff --git a/nova/virt/xenapi/fake.py b/nova/virt/xenapi/fake.py
index 72008a69d..050b696f0 100644
--- a/nova/virt/xenapi/fake.py
+++ b/nova/virt/xenapi/fake.py
@@ -536,8 +536,18 @@ class SessionBase(object):
 
     VDI_resize = VDI_resize_online
 
+    def _VM_reboot(self, session, vm_ref):
+        db_ref = _db_content['VM'][vm_ref]
+        if db_ref['power_state'] != 'Running':
+            raise Failure(['VM_BAD_POWER_STATE',
+                'fake-opaque-ref', db_ref['power_state'].lower(), 'halted'])
+        db_ref['power_state'] = 'Running'
+
     def VM_clean_reboot(self, session, vm_ref):
-        pass
+        return self._VM_reboot(session, vm_ref)
+
+    def VM_hard_reboot(self, session, vm_ref):
+        return self._VM_reboot(session, vm_ref)
 
     def VM_hard_shutdown(self, session, vm_ref):
         db_ref = _db_content['VM'][vm_ref]
diff --git a/nova/virt/xenapi/vmops.py b/nova/virt/xenapi/vmops.py
index 6af2ac9e0..5d445e0a6 100644
--- a/nova/virt/xenapi/vmops.py
+++ b/nova/virt/xenapi/vmops.py
@@ -843,10 +843,20 @@ class VMOps(object):
         # remove existing filters
         vm_ref = self._get_vm_opaque_ref(instance)
 
-        if reboot_type == "HARD":
-            self._session.call_xenapi('VM.hard_reboot', vm_ref)
-        else:
-            self._session.call_xenapi('VM.clean_reboot', vm_ref)
+        try:
+            if reboot_type == "HARD":
+                self._session.call_xenapi('VM.hard_reboot', vm_ref)
+            else:
+                self._session.call_xenapi('VM.clean_reboot', vm_ref)
+        except self._session.XenAPI.Failure, exc:
+            details = exc.details
+            if (details[0] == 'VM_BAD_POWER_STATE' and
+                    details[-1] == 'halted'):
+                LOG.info(_("Starting halted instance found during reboot"),
+                    instance=instance)
+                self._session.call_xenapi('VM.start', vm_ref, False, False)
+                return
+            raise
 
     def _get_agent_version(self, instance):
         """Get the version of the agent running on the VM instance."""